L1 learning consists not only in acquiring phonological, grammatical, lexical, etc. prescriptions of a Saussurean ‘langue’ as a system of signs, but also in subconscious inferences concerning prototypical usage and creativity measures of linguistic signs actually used by adult native speakers in communication. L2 acquisition by adults differs from L1 acquisition due to practical limitations. Multilingual text corpora are helpful as a means of overcoming this deficit of information. Non-native speakers relying on statistics of the right-hand and left-hand contexts of lexical items, and not only on frequency of separate lexical items, can measure the degree of creativity of original texts and of their own linguistic production in a foreign language. Thus, the English phrase ‘highly likely’ is used in English texts rather differently from the way it is cited and understood outside English texts proper, e. g. in Russian mass media. The paper analyzes statistics and semantics of the contexts of the English expression ‘highly likely’ and its synonyms in a big English corpus of original texts. Thus, in my English corpus, ‘highly likely’ is less frequent than ‘highly unlikely’, which signals negative rather than positive expectations of the propositions supplied with such hedges. These and similar data are compared with the use of their correspondences in a big Russian corpus of original texts. Pragmatic and semantic differences of these expressions in Russian and in English have to do with expectations, connotations and ‘implied inferences’ not always immediately perceived by Russian speakers citing the fashionable English phrase in their Russian discourse. The Russian transliteration of the English phrase, and not the literal translation, may be the only adequate way of rendering irony in trans-cultural communication.